Increasingly, it is common for people to communicate using multiple languages, often in a same communication, such as a text message. By way of example, in China, English terminology is often interleaved with Mandarin script. In such circumstances, use of existing monolingual language models is not sufficient to effectively implement language identification and word prediction. Further, given the scale of many monolingual language models, training and/or implementing a single multilingual language model exceeds the computational capabilities of contemporary systems.